Skip to content

Repository files navigation

InSession

A desktop app for therapists and coaches to record, transcribe, and query their client sessions — entirely on-device using local AI models.

InSession records audio from your microphone, transcribes it in real-time, stores the transcript, and lets you search across all sessions using natural language. No audio or text ever leaves your machine.

Features

  • 🎙️ Real-time transcription — live speech-to-text powered by NVIDIA Nemotron Speech
  • 💾 Session storage — transcripts saved locally as JSON
  • 🔍 Semantic search — query across all sessions using natural language (RAG with local embeddings and LLM)
  • 🔒 Fully local — all AI inference runs on-device via Foundry Local

Models

Purpose Model
Speech recognition nemotron-speech-streaming-en-0.6b
Embeddings (RAG) qwen3-embedding-0.6b
Chat / Q&A qwen2.5-1.5b

Models are downloaded automatically on first launch via Foundry Local.

Prerequisites

Getting Started

npm install
npm run tauri dev

To build a distributable app:

npm run tauri build

How It Works

  1. Record — Enter a client name and click Record. The app captures your microphone and streams audio to the Nemotron speech model for live transcription.
  2. Stop — Click Stop to end the session. The transcript is saved and automatically embedded for search.
  3. Search — Switch to the Search tab and ask a question. The app finds the most relevant session excerpts and uses the local LLM to generate an answer with source citations.

Tech Stack

  • Foundry Local SDK — local AI model management and inference
  • Tauri v2 — Rust + WebView desktop app framework
  • cpal — cross-platform audio capture
  • TypeScript + Vite (frontend)

Recommended IDE Setup

License

MIT — see LICENSE.

About

Therapy session transcriber and progress notes assistant

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages